Proactive environmental control of storage and exhibition facilities for cultural heritage through machine learning

Management of exhibition and storage environments is a fundamental collection care task at archives, libraries, museums and other cultural heritage institutions. Heritage institutions have various types of collections, with different value and use patterns, and their resources vary. Within this context, and whether the institution has simple or more advanced means of controlling the environment, the goal is to provide an indoor environment that optimizes the durability of the artefacts in the most sustainable way. To control and optimize the environment, collection managers need to understand the current storage conditions and it is therefore best practice to monitor indoor conditions regularly, including parameters such as temperature, relative humidity and air pollutants, as well as measures of the facility’s energy consumption. Today, control and optimization of the environment is retrospective, triggered by environmental events when they have already taken place. This means that mitigation actions may be initiated too late, when damage to cultural heritage has occurred. What if we could predict adverse environmental events and adjust the conditions before the collections are harmed? This paper explores the possibilities for developing a new proactive system for controlling indoor environmental conditions that will allow collection managers to act in a timelier manner to prevent damage to cultural heritage. Based on historical indoor environmental data as well as outdoor environmental data, the system applies machine learning algorithms to develop an early warning system for controlling indoor climate. To test our hypothesis we are currently conducting a case study, at a rurally located storage facility from 1990 housing a mixed collection of artefacts belonging to the National Museum of Denmark. The facility is a large purpose-built storage building (1,500 m2) and climatically controlled by semi-passive means (mainly by a thermally well-insulated building envelope, a low ventilation rate, and aided by mechanical dehumidification). We have plus ten years of historical indoor climate data from the facility, as well as more intermittently pollution data and energy consumption data. Besides, we have recently installed continuous recording instruments to measure air pollutants inside the facility and outdoors. In addition, we are collecting outdoor historic climate data, from a nearby metrological station owned by the Danish Meteorological Institute. We are experimenting with time series specific exploratory data analysis and combining this with domain knowledge, and exploring different approaches to detect patterns that should result in early warnings. We have started with establishing a baseline and are working with time series classification, prediction, and anomaly detection, as well as deep learning. The case study will be concluded summer 2021.